Clostridium novyi alpha-toxin-catalyzed incorporation of GlcNAc into Rho subfamily proteins.

The lethal and edema-inducing α-toxin from Clostridium novyi causes rounding up of cultured cell lines by redistribution of the actin cytoskeleton. α-Toxin belongs to the family of large clostridial cytotoxins that encompasses Clostridium difficile toxin A and B and the lethal toxin from Clostridium sordellii. Toxin A and toxin B have been recently identified as monoglucosyltransferases to modify the low molecular mass GTPases of the Rho subfamily (Just, I., Selzer, J., Wilm, M., Von Eichel-Streiber, C., Mann, M., and Aktories, K. (1995) Nature 375, 500-503 and Just, I., Wilm, M., Selzer, J., Rex, G., Von Eichel-Streiber, C., Mann, M., and Aktories, K. (1995) J. Biol. Chem. 270, 13932-13936). We report here the identification of the α-toxin-catalyzed modification of Rho. Using electrospray mass spectrometry, the mass of the modification was determined as 203 Da, consistent with a N-acetyl-hexosamine moiety. UDP-N-acetyl-glucosamine selectively served as cosubstrate for α-toxin-catalyzed modification into the Rho subfamily proteins Rho, Rac, Cdc42, and RhoG. The acceptor amino acid of N-acetyl-glucosaminylation was identified by mutagenesis as Thr-37 in Rho (equivalent to Thr-35 in Rac/Cdc42), which is located in the effector domain of the GTPases. C. novyi α-toxin seems to mediate its cytotoxic effects on cells by mimicking endogenous post-translational modification of cellular proteins.

Clostridium novyi type A strains have been identified as the causative organisms of gas gangrene infections of humans and animals (1). Type A strains produce an exotoxin, termed ␣-toxin, that exhibits in vivo both lethal and edematizing activity (1). In tissue culture, ␣-toxin is cytotoxic, causing cell shape changes that are accompanied by disruption of the microfilament cytoskeleton and by minor effects at the vimentin and tubulin system (2)(3)(4). Recently, ␣-toxin has been cloned and sequenced (5). As deduced from these data, ␣-toxin has a molecular mass of 250,166 Da and shows 48% homology with Clostridium difficile toxin A (ToxA) 1 and toxin B (ToxB). ToxA and ToxB are the major virulence factors of pathogenic C. difficile strains and have been identified as the causative agents of the antibiotic-associated diarrhea and the fatal form, the pseudomembranous colitis (6,7). Furthermore, ␣-toxin shows 34% homology to the lethal toxin from Clostridium sordellii (5,8), which is causally involved in diarrhea and enterotoxaemia in domestic animals and gas gangrene in man (1,9). These clostridial toxins share common structural features. The C-terminal part of the single-chained toxins covers repetitive peptides that are most likely involved in cell receptor binding, followed by a small hydrophobic intermediate region which probably participates in the translocation of the toxins into the cytoplasm of the target cell (5,10). The N-terminal part carries the biological activity (11). The common property of these intracellularly acting protein toxins is their cytotoxic activity, which leads to preferential destruction of the microfilament system of cell monolayers.
Recently, C. difficile ToxA and ToxB have been identified as monoglucosyltransferases that selectively modify the low molecular mass GTP-binding proteins of the Rho subfamily (12,13). The target proteins Rho, Rac, and Cdc42 are involved in the regulation of the actin cytoskeleton. Whereas Rho controls the formation of focal adhesions and stress fibers (14), Rac participates in membrane ruffling (15) and Cdc42 in formation of filopodia (16,17). Furthermore, the Rho subfamily proteins have been identified as being involved in the activation of transcription factors via the Ras-regulated pathway (18) and via a Ras-independent signal cascade (19 -21).
Here we report the identification of C. novyi ␣-toxin as a N-acetyl-glucosaminyltransferase that modifies the Rho subtype proteins). 2 (24) and C. difficile toxin B (25) were purified as described.
Cell Culture-NIH 3T3 cells were grown in Dulbecco's medium supplemented with 10% fetal calf serum, 4 mM glutamine/penicillin/streptomycin. After 24 h the medium was changed, and cells were incubated with the toxins for the indicated times. Before cell lysis, the cells were rinsed with ice-cold phosphate-buffered saline, pH 7.2, and were then disrupted mechanically by sonication (five times on ice) in the presence of lysis buffer (2 mM MgCl 2 , 40 g/ml aprotinin, 0.3 mM phenylmethylsulfonyl fluoride, 20 g/ml leupeptin, 80 g/ml benzamidine, 50 mM HEPES, pH 7.4) followed by centrifugation for 10 min at 2,000 ϫ g. The supernatant was used for glycosylation reactions.
ADP-ribosylation of Rho in Lysates from Toxin-treated NIH 3T3 Cells-NIH 3T3 cells (2 ϫ 10 6 cells/well) were incubated with ␣-toxin (400 ng/ml) for the indicated periods of time. Thereafter, the cells were lysed as described above, and the cell lysates were centrifuged for 60 min at 100,000 ϫ g. The pellets were dissolved in 2 mM MgCl 2 /1 mM dithiothreitol/0.3 mM phenylmethylsulfonyl fluoride/50 mM HEPES, pH 7.4, and used as membrane fraction. [ 32 P]ADP-ribosylation of the membrane fractions was carried out as described. The SDS-PAGE was evaluated with PhosphorImager SF (Molecular Dynamics). The amount of ADP-ribosylation was calculated as the percentage of control (nontreated cells).
Preparation of Recombinant GTP-binding Proteins-RhoA, RhoA T37A , RhoA P36R , Rac1, Cdc42, RhoG, and H-Ras were prepared from their fusion proteins (e.g. RhoA-glutathione S-transferase) as described (22). Glutathione S-transferase fusion proteins from the E. coli expression vector pGEX-2T were isolated by affinity purification with glutathione-Sepharose (Pharmacia Biotech Inc.) followed by cleavage of GTP-binding proteins from the glutathione S-transferase fusion protein by thrombin treatment (100 g/ml for 60 min at 22°C). Thrombin was removed by binding to benzamidine-Sepharose, and the GTP-binding proteins were concentrated with Centricon (Amicon).
Glucosylation Reaction-Recombinant GTP-binding proteins (50 g/ ml) or cell lysates were incubated with ␣-toxin (40 g/ml) in a buffer containing 30 M UDP-[ 14 C]GlcNAc/2 mM MgCl 2 /150 mM KCl/50 mM HEPES, pH 7.4, for the indicated times at 37°C. For guanine nucleotide exchange Cdc42 was incubated with 300 M of GDP, GTP, or GTP␥S in the presence of 5 mM EDTA for 3 min at 22°C followed by the addition of 7 mM MgCl 2 .
Cytosolic Subfractions (F3000/F500)-Rat brains were homogenized in three volumes of lysis buffer and centrifuged for 60 min at 100,000 ϫ g. The supernatant (cytosolic fraction) was incubated for 15 min at 95°C, and denatured proteins were removed by centrifugation. The supernatant was passed through an ultrafiltration membrane (Amicon Corp.) with 3000 and 500 Da cut-offs, respectively. The flow-through termed F3000 or F500 was used as cytosolic subfraction.
Gel Electrophoresis-Proteins were dissolved in sample buffer and subjected to 12.5% SDS-PAGE (27) followed by analysis with the Phos-phorImager from Molecular Dynamics.
Purification of Toxin-treated RhoA-RhoA (100 g) was treated without or with ␣-toxin (40 g/ml) in the presence of the cytosolic subfraction F3000 for 90 min at 37°C and was then separated by high pressure liquid chromatography on Sephasil RP-C18 column (Pharmacia). The proteins were eluted with a linear gradient from 0.1% trifluoroacetic acid in water to 0.1% trifluoroacetic acid and 70% acetonitrile. RhoA eluting at about 70% of the gradient was freeze dried.
Mass Spectrometry-To determine the molecular mass of the intact protein purified RhoA was subjected to electrospray mass spectrometry (28,29) using the nano-electrospray source (30,31). Determination of the molecular mass was performed with the Sciex API III triple quadrupole mass spectrometer from Sciex-Perkin-Elmer.

RESULTS AND DISCUSSION
Identification of the Modification-To test whether ␣-toxin also exhibits glucosyltransferase activity as do C. difficile ToxA and ToxB, lysates from NIH 3T3 cells and recombinant Rho subtype proteins, respectively, were incubated with ␣-toxin in the presence of UDP-[ 14 C]glucose. Surprisingly, under conditions where ToxA/B elicited full glucosyltransferase activity, ␣-toxin did not catalyze incorporation of glucose from UDPglucose into cellular or recombinant proteins (data not shown). Therefore, we first tested whether the Rho protein is actually the target of ␣-toxin. To this end, ADP-ribosyltransferase C3 was used to detect alterations in the Rho protein. C3 selectively ADP-ribosylates the Rho subtype proteins RhoA, B, and C in Asn-41, thereby inactivating Rho (34 -37). In lysates prepared from ␣-toxin-treated NIH 3T3 cells, C3-catalyzed [ 32 P]ADPribosylation of Rho was decreased in a time-dependent manner comparable with the effects of ToxA and ToxB on cellular Rho (Fig. 1A). Heat inactivation of ␣-toxin abolished both cytotoxic effects on cells and inhibition of Rho-ADP-ribosylation (not shown), indicating a specific effect of ␣-toxin on the ADPribosylation of Rho. As observed with ToxA/B, ␣-toxin affected ADP-ribosylation of recombinant Rho protein exclusively in the presence of a cytosolic fraction indicating a cytosolic cofactor is essential for ␣-toxin activity (Fig. 1B). This cytosolic cofactor was partially characterized as a heat-stable, nonproteinacious agent with a molecular mass between 500 and 3000 Da. Partial purification of the cofactor was essential to produce sufficient amounts of ␣-toxin-modified recombinant Rho, which was subsequently subjected to electrospray mass spectrometry to determine the molecular mass of the modification. The mass of the complete Rho protein modified by ␣-toxin was 203 Da higher than the unmodified protein (not shown), suggesting that modified Rho contains a covalently bound N-acetyl-hexosamine moiety (221 Da of N-acetyl-hexosamine minus 18 Da of released H 2 O). These data are consistent with monoglycosylation of Rho.
The identification of the type of hexosamine was performed with biochemical methods by testing several sugars to inhibit ␣-toxin-induced decrease in Rho ADP-ribosylation. Because the molecular mass of the cosubstrate was Ͼ500 Da, mere Nacetyl-hexosamines were excluded, and the activated forms (UDP-N-acetyl-hexosamines) were tested. As shown in Fig. 2, ␣-toxin induced inhibition of C3-catalyzed ADP-ribosylation of Rho exclusively in the presence of UDP-GlcNAc. Other UDPhexoses did not mediate this effect. To corroborate the incorporation of GlcNAc directly, [ 14 C]labeled nucleotide sugars were used. As shown in Fig. 3A, ␣-toxin catalyzed incorporation of [ 14 C]GlcNAc but not of [ 14 C]Glc or [ 14 C]GalNAc into Rho. Denaturing of either Rho or ␣-toxin resulted in complete inhibition of modification consistent with the notion that the native protein structure is essential for this type of glycosylation. Taken together, these data indicate that C. novyi ␣-toxin is a glycosyltransferase that utilizes the cosubstrate UDP-GlcNAc to transfer the GlcNAc moiety to the Rho protein.
Substrate Specificity-To identify the substrate proteins of ␣-toxin, several recombinant low molecular mass GTPases were tested. As illustrated in Fig. 3B, Rho, Rac, Cdc42, and RhoG, all members of the Rho subfamily, were N-acetyl-glucosaminylated, whereas other GTPases of the Ras superfamily namely H-Ras, Arf1, Rab5, and Ran were not target for ␣-toxin.
N-Acetylglucosaminylation is significantly stimulated in the presence of KCl with maximal effects at 150 mM, whereas NaCl had no stimulatory effect. Thus, ␣-toxin modifies the same recombinant substrate proteins as do ToxA and ToxB from C. difficile.
Acceptor Amino Acid-To test whether ␣-toxin uses the same acceptor site as ToxA and ToxB, sequential glycosylation was performed. Modification of RhoA with ␣-toxin in the presence of unlabeled UDP-GlcNAc, followed by a second glycosylation in the presence of UDP-[ 14 C]Glc and ToxB resulted in blocked incorporation of [ 14 C]Glc (Fig. 4). The same was true when the first glycosylation was performed with ToxB and UDP-Glc. These results indicate that ␣-toxin shares the same acceptor amino acid in RhoA, namely Thr-37. To prove the acceptor amino acid Thr-37 by a different approach, we tested whether exchange of Thr in position 37 for Ala abolishes incorporation of GlcNAc (Fig. 4). RhoA T37A was still substrate for ADP-ribosyltransferase C3, indicating no gross changes in the overall protein structure. Incorporation of GlcNAc by ␣-toxin, however, was completely blocked. Thus, Thr-37 in Rho is the acceptor amino acid, and there are no alternative acceptor sites. Exchange of the preceding amino acid Pro to Ala (RhoA P36R ) decreased but did not completely blocked modifcation. As can be deduced from the primary structure of the Rho subfamily proteins, Thr-37 in RhoA corresponds to Thr-35 in Rac and Cdc42. ␣-Toxin shares the acceptor amino acid in the Rho subfamily proteins with ToxA and ToxB (12,13).
Applying the crystal structure of H-Ras to Cdc42, Thr-35 of Cdc42 (equivalent to Thr-35 in H-Ras) is located in the GTPbinding and hydrolyzing domain (G-2). The hydroxyl group of Thr-35 is ligand for the Mg 2ϩ ion, which is involved in the coordination of the ␤-/␥-phosphates of the guanyl nucleotide (38). Hydrolysis of the ␥-phosphate induces movement of loop L2 resulting in rearrangement of Thr-35. The hydroxyl group of Thr-35 is exposed at the surface of Cdc42 and is now accessible for glucosylation. Consistent with this model is the finding that Cdc42 is a better substrate for ␣-toxin in the GDP-bound form than in the GTP-bound form (Fig. 5). Binding of the nonhydrolyzable GTP␥S almost completely blocked modification of Cdc42, suggesting that incorporation of GlcNAc into GTPbound Cdc42 is due to GTP hydrolysis. Coordination of the ␥-phosphate results in a conformation of Thr-35, which is incompatible with attachment of GlcNAc. In the GDP-bound form incorporation did not exceed 1 mol of GlcNAc/mol of Cdc42, consistent with monoglycosylation of Cdc42. These data are in agreement with the data of ToxA and ToxB, which catalyze monoglucosylation (12,13).
Thr-37 in Rho (Thr-35 in Cdc42 and Rac) is located in the effector domain of these GTPases. In case of Rac and Cdc42, the p65 PAK kinase has been identified as an effector, whereas the effector of the Rho subtype protein appears to be a p150 serine/ threonine kinase (39). In Rac two different effector sites have been identified to interact with the serine/threonine p65 PAK kinase and p67 phox (40). The N-terminal site covers residues 22-45 embracing the acceptor of GlcNAc, Thr-35. It is conceivable that the hydrophilic GlcNAc moiety in this crucial domain impedes with effector coupling resulting in blocked signal cascade.
In Vivo Substrate Proteins-Incubation of cell lysates with ␣-toxin and UDP-[ 14 C]GlcNAc resulted in labeling of two protein bands (Fig. 6). Immunoblot analysis of gel electrophoretically separated cellular proteins with anti-RhoA and anti-Cdc42, respectively, showed that the lower band corresponds to Cdc42 and the upper one to RhoA (data not shown). In contrast to ToxA/B, ␣-toxin catalyzed only a faint labeling of Rho (upper band). It seems that cellular Rho is a poor substrate for ␣-toxin. However, this observation is not unique to NIH 3T3 cells.
Lysates from various cultured cell lines showed a low extent in N-acetyl-glucosaminylation of Rho compared with Cdc42/Rac. To test whether isoprenylation is the basis for this observation NIH 3T3 cells were treated with lovastatin (30 M for 24 h) to block isoprenylation. However, ␣-toxin-catalyzed incorporation of GlcNAc into Rho from this lysates did not increase (data not shown). Furthermore, proteolytical degradation of Rho was excluded to cause low incorporation of GlcNAc into Rho. Thus, for unknown reasons recombinant Rho is a superior substrate to cellular Rho.
To test whether Cdc42 and Rac were modified by ␣-toxin in the intact cell, the method of [ 14 C]galactosylation of proteins bearing a GlcNAc moiety was employed (32). Lysates from ␣-toxin-treated NIH 3T3 cells were denatured by incubation at 95°C in the presence of detergents followed by incubation with UDP-[ 14 C]Gal and galactosyltransferase from bovine milk. As demonstrated in Fig. 7A, [ 14 C]galactosylated proteins in the molecular mass range of 20 -22 kDa were only detected in lysates from ␣-toxin-treated cells. Identification of the [ 14 C]galactosylated protein was performed in a second approach. Cell lysates were electrotransferred to nitrocellulose and the blot slice corresponding to the molecular mass range of 15-25 kDa was cut out and was subjected to the [ 14 C]galactosylation reaction (Fig. 7B). Thereafter, the same blot slice was probed with anti-Cdc42 identifying the labeled protein as Cdc42 (Fig. 7B). These data suggest that ␣-toxin modifies Cdc42 in intact cells.
O-linked N-acetylglucosaminylation is not unique to C. novyi ␣-toxin, but it is a common post-translational modification in eukaryotic cells. Characteristic for O-GlcNAc-bearing proteins is their intracellular localization in the cytoplasm and in the nuclear space, respectively (41). This kind of modification is as highly dynamic as phoshorylation (41). The function of monosaccharide attachment is not fully understood, but there is evidence that O-glycosylation with GlcNAc is involved in nuclear targeting and formation of multimeric protein structures (41,42). It is conceivable that bacterial toxins mimic regulatory mechanisms of eukaryotic cells to interfere effectively with the signal cascade of their target cells. C. novyi ␣-toxin seems to belong to those bacterial toxins that mediate their cytotoxic effects by mimicking endogenous post-translational modification of cellular proteins as has been reported for diphtheria toxin/pseudomonas exotoxin A (43) and for cholera toxin (44) that mimic endogenous ADP-ribosylation.
In conclusion, ␣-toxin from C. novyi has been identified as a monoglycosyltransferase that catalyzes incorporation of Glc-NAc into cellular Rho, Cdc42, and Rac. Modification of Thr-35 in Cdc42/Rac causes inactivation of these GTPases resulting in redistribution of the actin cytoskeleton. In common with C. difficile ToxA and ToxB, the protein substrate specificity of ␣-toxin is confined to the Rho subfamily proteins. In contrast to ToxA and ToxB, which use UDP-glucose as cosubstrate, ␣-toxin transfers a N-acetyl-glucosamine moiety from the activated UDP precursor to the target proteins.